# Obtaining modern CUDA libraries

**URL:** <https://discourse.julialang.org/t/obtaining-modern-cuda-libraries/123237>\
**Category:** General Usage\
**Created:** [November 29, 2024, 8:07am UTC](https://discourse.julialang.org/t/obtaining-modern-cuda-libraries/123237 "2024-11-29T08:07:45Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![jw3126](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/jw3126/32/3086_2.png) [@jw3126](https://discourse.julialang.org/u/jw3126)\
**Post date:** [November 29, 2024, 8:07am UTC](https://discourse.julialang.org/t/obtaining-modern-cuda-libraries/123237/1 "2024-11-29T08:07:45Z")

</div>

For [ONNXRunTime.jl](https://github.com/jw3126/ONNXRunTime.jl) GPU support, we need various CUDA libraries. In theory, we only need binaries.  
In practice, the easiest way to obtain these was to install the following packages:

- [CUDA.jl](https://github.com/JuliaGPU/CUDA.jl) and [CUDNN](https://github.com/JuliaAttic/CUDNN.jl)

However CUDNN is deprecated now and this causes trouble. For instance, it forces us to choose between a modern [onnxruntime and onnx GPU support](https://github.com/jw3126/ONNXRunTime.jl/pull/40).

What would be a good way to obtain modern compatible cuda binaries? Especially libcudnn?

---

<div class="post-metadata">

**Author:** ![GunnarFarneback](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/gunnarfarneback/32/1827_2.png) [@GunnarFarneback](https://discourse.julialang.org/u/GunnarFarneback)\
**Post date:** [November 29, 2024, 11:16am UTC](https://discourse.julialang.org/t/obtaining-modern-cuda-libraries/123237/2 "2024-11-29T11:16:36Z")

</div>

> However CUDNN is deprecated

For the record this was a misunderstanding based on the extremely out of date [GitHub - JuliaAttic/CUDNN.jl: Julia wrapper for the NVIDIA cuDNN GPU deep learning library](https://github.com/JuliaAttic/CUDNN.jl).